How to use data to design the best market for your company

Your business spends a lot of time analyzing the market you sell to, but can you take a step back and instead create the market that you want? Experts in the little-known field of behavioral economics and market design say “yes.”

“Market design says ‘we don’t always get the outcomes we want so let’s tweak the markets and yet still harness market forces’,” says R Preston McAfee, Microsoft’s chief economist. It’s a set of principles that explains market behavior and allows us to predict and even change it, with incentive and other tools.

[ Also on CIO.com: Docker, machine learning are top tech trends for 2017 ]

“It bears the same relationship to traditional economics that engineering bears to physics,” McAfee explains. It’s been around since the 1980s and it’s used for assigning physicians to medical residencies as well as for spectrum auctions (which he advised the FCC on), with about $100 billion spent through each of those markets. “It’s accepted technology, but it’s almost unheard of.”

That’s changing. Google and Amazon also have chief economists (and McAfee himself previously held similar roles at Google and Yahoo). Market design could be the latest prohibitively expensive discipline to be automated and go mainstream, he suggests. “Traditionally, you hired a brilliant person who did a handcrafted market design. This is about to change, and it’s about to change because of big data. Sensors, data and machine learning allow an entirely new type of market design, the automation of market design.”

[ Also on CIO.com: Download CIO’s guide to understanding analytics ]

The first signs of that are services like Uber and AirBnB, says McAfee. “They’ve hired market designers because they’re in the business of markets. Networks are always markets; you’re connecting buyers and sellers. Market design helps make them even more valuable as we harness this potential. We’re on the verge of a revolutionary period in market design where we start seeing machine-run, machine-led, data-driven market design.”

Governing a market

Designing a market means taking on all the activities of a government, McAfee suggests. “Take Walmart, they have suppliers who sell independently through Walmart, much like Amazon Marketplace. They face a market problem: Walmart stands in the middle of the transaction between suppliers and buyers. A business should think of itself as the controller and government of that market. It needs to provide the equivalent of infrastructure for a government, a physical or virtual place where transactions take place. It wants to understand its citizens, the buyers and sellers: what are their interests, how are they going to respond? What will lubricate the systems and make these markets more efficient? And then they should build the tools that make it more efficient.”

That might be the button that Amazon offers to “subscribe” to a product you order frequently, or the smart shopping cart that J&J Foods built in Dynamics using the Cortana Intelligence suite to put the products customers are most likely to want in their carts automatically (a move that has increased revenue around 5 percent).

Like a government, markets face competition. “Walmart has a market, Amazon has a market and the consumer can go to either, the same way the U.S. and U.K. governments have to worry that my company might incorporate in Ireland … because another thing markets can do is offer subsidies and impose taxes.”

[ Also on CIO.com: Comparing AI tools in Salesforce Einstein and Dynamics 365 ]

It’s important to target those incentives at the right customers. “One of the principles of market design is you want to give incentives to the people who can make a difference and at the place where they can make a difference. If you give me an incentive, but I don’t have the tools to make a difference it’s a waste, it just imposes risk on me that I can’t do anything about.”

Start by knowing it’s a market

Understanding that your business is building a market and that you have a responsibility to make it work efficiently is rare — even among major players like Apple. “When Apple launched its innovative App Store it didn’t realize it faced a market design problem, and as a result it just sold the apps,” McAfee says. He says that offering subscriptions so that people “rent” apps rather than buying them would have improved the quality of apps — even though it would have meant fewer apps overall. “If you want to grow an app store really fast, if you want to reach a million apps really fast, sell them — and you’ll have hundreds of fart apps. If you rent them, what you get is a relatively small number of really good apps. So from a social perspective, it’s better to rent them.”

“That’s an example of a market design failure that could have been more successful. They said ‘let’s make a market’ without any apparent thought about what are the different tools. By not thinking about the App Store as a market they made several mistakes that Google didn’t make [when they designed their ad market], because Google thought about it more seriously as running a market.”

Building security in

Looking at behavior produces incentives that sound very much like security best practices. “Most companies are still trying to use perimeter defenses; the correct thing to do is actually interior defenses. If I’ve never asked for my CEO’s files and I come along and ask for those files, that ought to get escalated. I may have to do five-factor authentication to get access to files I’m legitimately allowed to see but have never asked for; on the other hand if this is a SharePoint site I’ve been to every day for the last two and a half years I might not even need a password if I’m on the same IP address.”

But behavior is only one of the tools available. Take the problem of ‘flash crashes’ in automated trading, which are likely to get worse since more than half the volume of the New York Stock Exchange is now automated. “There is no reason the exchange couldn’t have some protections built in. Simple market designs could solve many of the problems,” McAfee says.

To deal with flash crashes, he suggests running transactions in batches, 10 times a second; that adds a little latency. “The exchange itself becomes automated — its protection, its design, its feedback loops. Under normal circumstances, it might take a very light touch and not interfere with trades very much but when things start to get unexpected and volatility goes up, it starts dampening behavior. It starts injecting larger buffers into the system, so we can protect against a bad feedback loop that spills over on itself.”

There are several other areas we could treat as markets, including traffic, parking and crowd control. “We can forecast what traffic patterns are going to look like, so maybe I start getting directed to different places, or I’m told I can pay less to drive on this road if I wait 15 minutes. All of this is about incentives; it’s all about getting the right behavior,” he says.

Israel has recently experimented with this kind of road pricing models to ease congestion: 46 percent of drivers in the areas changed their driving patterns as a result. One company, adele:systems, has some ambitious plans to unify tolling across Europe with a system that can handle charges based on mileage, time on the road, the combination of weight and distance, the number of axles on a truck or driving into specific areas like downtown city areas. That way, trucks won’t divert onto residential streets to avoid the tools on highways. It’s important not to accidentally incentivize bad behavior.

Once you start treating situations like a network system, you can look for the points when you need to intervene. And increasingly, McAfee says, you can start automating that. Some of that could be simple, like agricultural systems that only water and apply fertilizer when they’re needed instead of on a fixed schedule. “This is cheap technology; the payback period is a few months so one crop rotation easily pays back for the technology,” McAfee claims.

Privacy and false positives

Key to automating these complex systems that involve people is understanding human behavior and motivation. “I think we are much simpler than we expect and much more predictable, and we’re just now reaching computational power to uncover those patterns,” McAfee says.

“We’re coming to a point where machine understanding of human capabilities is going to be really transformational; I don’t think we really appreciate just how fundamental this is.”

A caveat, however, is that businesses are going to have to understand a lot more about the data they need to be able to design markets. “There are three different buckets of how transitory and how valuable data is,” McAfee says. Some data depreciates in value very quickly, like daily browsing data. Some data is valuable but takes time to gather. For example, “if you investigate ski resorts every fall, having years of data is quite valuable; it tells me not just what to show you but when to show you. But you can’t do that without long enough series data to see that pattern.” Other data is valuable immediately and in the longer term; “these are areas where my interest never fades.”

For that data to be really valuable, you have to know what it correlates with. “I find myself explaining over and over again that it doesn’t matter how much data you have if you don’t have ground truth. If I want to predict health outcomes, I need a population of people whose health outcomes I already know.”

Another caveat: The more data you have the more likely it becomes that predictions and correlations will be incorrect. “This problem of spurious correlation gets bigger the bigger the data gets,” McAfee warns. “We have not really prepared ourselves for all the failings of machine learning that are in our future.”

To avoid a backlash, he suggests that systems should only offer predictions that have a high certainty. “The problem we’ve solved is ‘which is the thing that’s ranked highest’, but what I care about is ‘is it relevant?’. Make the first thing you say be something you’re 99 percent sure is dead on relevant.” And if not? “Systems should be silent more often than they are,” he suggests.

McAfee can’t say how Microsoft might offer automated market design. Certainly, it would fit in well with the other analytics tools in the Cortana Intelligence Suite and the predictive intelligence tools in Dynamics. And it would make sense as a cloud service, as McAfee notes that the infrastructure to run predictive analytics often costs as much as the value it offers businesses.

Related video